Parameter-Masked Decoupled Optimization for Cross-Domain Class-Incremental Learning

Ziqi Gu, Chunyan Xu, Yangguang Liu, Wenxuan Fang, Baotong Su, Tong Zhang, Dan Wang, Zhen Cui
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:37292-37307, 2026.

Abstract

Cross-domain class-incremental learning (CD-CIL) requires models to continuously acquire new classes across shifting domains while retaining previously learned knowledge. Existing approaches often entangle what to update with how to update, resulting in unstable adaptation and severe forgetting under domain shifts. Inspired by the hippocampal learning mechanism that separates rapid adaptation from stable consolidation, we propose Parameter-Masked Decoupled Optimization (PMDO) that disentangles what knowledge is adapted from how learning proceeds in cross-domain class-incremental learning. We introduce a domain-aware knowledge decoupler that selectively adapts domain-relevant shared parameters, constraining incremental updates while preserving prior representations. To regulate how learning proceeds, we further design a stability-aware trajectory regulation that guides optimization along transferable and stable optimization trajectories, thereby reducing interference across domain transitions. PMDO enables effective cross-domain adaptation while mitigating catastrophic forgetting and maintaining long-term learnability. Extensive experiments across multiple benchmarks demonstrate the effectiveness of PMDO and its superiority over state-of-the-art methods.

Cite this Paper

BibTeX
@InProceedings{pmlr-v306-gu26n, title = {Parameter-Masked Decoupled Optimization for Cross-Domain Class-Incremental Learning}, author = {Gu, Ziqi and Xu, Chunyan and Liu, Yangguang and Fang, Wenxuan and Su, Baotong and Zhang, Tong and Wang, Dan and Cui, Zhen}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {37292--37307}, year = {2026}, editor = {Zhang, Tong and Dudik, Miroslav and Jaggi, Martin and Agarwal, Alekh and Li, Sharon and Schuurmans, Dale and Zhu, Jerry and Berkenkamp, Felix and Dong, Hanze and Bietti, Alberto}, volume = {306}, series = {Proceedings of Machine Learning Research}, month = {06--11 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v306/main/assets/gu26n/gu26n.pdf}, url = {https://proceedings.mlr.press/v306/gu26n.html}, abstract = {Cross-domain class-incremental learning (CD-CIL) requires models to continuously acquire new classes across shifting domains while retaining previously learned knowledge. Existing approaches often entangle what to update with how to update, resulting in unstable adaptation and severe forgetting under domain shifts. Inspired by the hippocampal learning mechanism that separates rapid adaptation from stable consolidation, we propose Parameter-Masked Decoupled Optimization (PMDO) that disentangles what knowledge is adapted from how learning proceeds in cross-domain class-incremental learning. We introduce a domain-aware knowledge decoupler that selectively adapts domain-relevant shared parameters, constraining incremental updates while preserving prior representations. To regulate how learning proceeds, we further design a stability-aware trajectory regulation that guides optimization along transferable and stable optimization trajectories, thereby reducing interference across domain transitions. PMDO enables effective cross-domain adaptation while mitigating catastrophic forgetting and maintaining long-term learnability. Extensive experiments across multiple benchmarks demonstrate the effectiveness of PMDO and its superiority over state-of-the-art methods.} }
Endnote
%0 Conference Paper %T Parameter-Masked Decoupled Optimization for Cross-Domain Class-Incremental Learning %A Ziqi Gu %A Chunyan Xu %A Yangguang Liu %A Wenxuan Fang %A Baotong Su %A Tong Zhang %A Dan Wang %A Zhen Cui %B Proceedings of the 43rd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2026 %E Tong Zhang %E Miroslav Dudik %E Martin Jaggi %E Alekh Agarwal %E Sharon Li %E Dale Schuurmans %E Jerry Zhu %E Felix Berkenkamp %E Hanze Dong %E Alberto Bietti %F pmlr-v306-gu26n %I PMLR %P 37292--37307 %U https://proceedings.mlr.press/v306/gu26n.html %V 306 %X Cross-domain class-incremental learning (CD-CIL) requires models to continuously acquire new classes across shifting domains while retaining previously learned knowledge. Existing approaches often entangle what to update with how to update, resulting in unstable adaptation and severe forgetting under domain shifts. Inspired by the hippocampal learning mechanism that separates rapid adaptation from stable consolidation, we propose Parameter-Masked Decoupled Optimization (PMDO) that disentangles what knowledge is adapted from how learning proceeds in cross-domain class-incremental learning. We introduce a domain-aware knowledge decoupler that selectively adapts domain-relevant shared parameters, constraining incremental updates while preserving prior representations. To regulate how learning proceeds, we further design a stability-aware trajectory regulation that guides optimization along transferable and stable optimization trajectories, thereby reducing interference across domain transitions. PMDO enables effective cross-domain adaptation while mitigating catastrophic forgetting and maintaining long-term learnability. Extensive experiments across multiple benchmarks demonstrate the effectiveness of PMDO and its superiority over state-of-the-art methods.
APA
Gu, Z., Xu, C., Liu, Y., Fang, W., Su, B., Zhang, T., Wang, D. & Cui, Z.. (2026). Parameter-Masked Decoupled Optimization for Cross-Domain Class-Incremental Learning. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:37292-37307 Available from https://proceedings.mlr.press/v306/gu26n.html.

Related Material